NVIDIAFlagship
HGX H200
Largest NVIDIA memory for LLM training and inference
- GPU memory
- 141 GB HBM3e
- Power
- 700 W
- Form factor
- SXM5 · NVLink
- Per server
- Up to 8
Max per server1.1 TB
Get pricing with this GPU
PowerEdge XE · 8-GPU AI server
Dell PowerEdge XE9680Dell's 8-GPU platform for training and serving large AI models.
Eight NVIDIA HGX H200 or H100 GPUs with NVLink, or eight AMD Instinct MI300X or Intel Gaudi 3 accelerators, with two 5th Gen Intel Xeon processors in an air-cooled 6U chassis. Sized, quoted and delivered by Forte Tech across Egypt and the GCC.
- Genuine Dell hardware
- Egypt & GCC delivery
- Local pre-sales engineers
- Tender & PO ready
8 × GPUNVLink, Infinity Fabric or RoCE
Air-cooled 6UNo liquid cooling required
8× GPU
NVIDIA H200 / H100, AMD MI300X or Intel Gaudi 3
1.5TB
GPU memory with 8 × AMD Instinct MI300X
4TB
DDR5 across 32 DIMMs at up to 5600 MT/s
10× Gen5
PCIe x16 slots for high-speed GPU networking
Built for
Built for enterprise and national-scale AI
One XE9680 trains, fine-tunes and serves large models. Several of them form the core of an on-premises AI cluster.
LLM training & fine-tuning
Train or fine-tune large language models on eight tightly connected GPUs.
Large-scale inference
Serve LLMs and RAG assistants to thousands of users with up to 1.5 TB of GPU memory.
HPC & simulation
Accelerate scientific computing, risk modelling and engineering simulation.
Sovereign & private AI
Build AI platforms for government, banking and telecom with data kept on-premises.
Accelerator options
Train and serve large models on‑premises
The XE9680 links eight accelerators over a high-bandwidth fabric, so a large model can run across all of them as one system, inside your own data center.
- 8-way accelerator fabric: NVIDIA NVLink on HGX H200 and H100, AMD Infinity Fabric on MI300X, or embedded RoCE Ethernet on Intel Gaudi 3.
- Up to 1.5 TB of GPU memory: room to train, fine-tune and serve large models inside a single server.
- Networking for AI clusters: up to 10 PCIe Gen5 x16 slots for GPU networking, or 6 × 800 Gb OSFP ports built in with Gaudi 3.
- Built for the data center: air-cooled 6U chassis, Titanium 2800 W or 3200 W power supplies and iDRAC9 management.
Private LLMs & RAG
Arabic and English assistants trained on your own documents, served from your data center.
Model training
Train and fine-tune foundation models on your own data.
Computer vision at scale
City-scale video analytics, document processing and inspection.
Analytics & HPC
Fraud detection, risk modelling and scientific workloads on GPUs.
Accelerator options for the XE9680
Every XE9680 uses eight of the same accelerator. Choose the platform and we'll quote a complete, validated system.
NVIDIAProven
HGX H100
LLM training, fine-tuning and HPC
- GPU memory
- 80 GB HBM3
- Power
- 700 W
- Form factor
- SXM5 · NVLink
- Per server
- Up to 8
Max per server640 GB
Get pricing with this GPU
AMDMost memory
Instinct MI300X
Very large models and memory-bound inference
- GPU memory
- 192 GB HBM3
- Power
- 750 W
- Form factor
- OAM · Infinity Fabric
- Per server
- Up to 8
Max per server1.5 TB
Get pricing with this GPU
IntelOpen Ethernet
Gaudi 3
Training and inference with built-in 800 Gb Ethernet
- GPU memory
- 128 GB HBM2e
- Power
- 900 W
- Form factor
- OAM · RoCE
- Per server
- Up to 8
Max per server1 TB
Get pricing with this GPU
Not sure which GPU?
Tell us your model, data and number of users. Our engineers will recommend the right GPU and a complete, compatible build.
Ask our engineersSizing rule of thumb: inference needs about 2 GB of GPU memory per billion parameters at FP16, or about 1 GB at FP8. Full training needs several times more. For example, a 405-billion-parameter model at FP8 (about 405 GB) fits on 8 × H200 (1.1 TB).
Deployment: Dell requires ProDeploy Plus or Dell Customer Deployment services for XE-series servers, including GPU subsystem testing. We include the required deployment and review your site's power, cooling and floor loading before delivery.
What's inside
Engineered for 8-GPU workloads
The XE9680 pairs eight data-center accelerators with Intel Xeon processors, DDR5 and PCIe Gen5 networking in one air-cooled system.
Accelerators
8 × SXM5 or OAM
NVIDIA HGX H200 or H100 with NVLink, AMD Instinct MI300X with Infinity Fabric, or Intel Gaudi 3 with RoCE.
Compute
Two 5th Gen Intel Xeon
Up to 64 cores per processor (or 4th Gen with up to 56) to keep the GPUs fed with data.
Memory
4 TB DDR5
32 RDIMM slots at up to 5600 MT/s for data pipelines and pre-processing.
Storage
Up to 16 × E3.S NVMe
Or 8 × 2.5" NVMe/SAS/SATA (122.88 TB), with BOSS-N1 mirrored boot.
Networking
Up to 10 × PCIe Gen5 x16
For high-speed InfiniBand or Ethernet GPU networking, plus OCP 3.0 and 2 × 1 GbE.
Security & management
Silicon root of trust + iDRAC9
Signed firmware, Secure Boot and TPM 2.0, managed with iDRAC9, Redfish and OpenManage Enterprise.
Popular configurations
Start from a reference system
Reference configurations our engineers adjust to your models, users, networking and data-center limits.
H200 Training Node
Eight NVIDIA H200 GPUs for training, fine-tuning and large-model inference.
- 2 × 5th Gen Intel Xeon Platinum CPUs
- 2 TB DDR5 RDIMM
- 8 × NVIDIA H200 141 GB SXM5 (NVLink)
- 8 × E3.S NVMe for datasets and checkpoints
- 8 × 400 Gb GPU networking
- 2800 W Titanium PSUs
MI300X Inference Node
1.5 TB of GPU memory for very large models and high-concurrency inference.
- 2 × 5th Gen Intel Xeon Platinum CPUs
- 2 TB DDR5 RDIMM
- 8 × AMD Instinct MI300X 192 GB (OAM)
- 8 × E3.S NVMe
- 8 × 400 Gb networking
- 2800 W Titanium PSUs
Gaudi 3 AI Node
Intel Gaudi 3 accelerators with built-in 800 Gb Ethernet for scale-out AI.
- 2 × 5th Gen Intel Xeon Platinum CPUs
- 2 TB DDR5 RDIMM
- 8 × Intel Gaudi 3 128 GB (OAM)
- 6 × 800 Gb OSFP (built in)
- 8 × 2.5" NVMe SSD
- 3000 W Titanium PSUs
Need something different? Send us your requirements or BOM and our engineers will propose the right build.
Technical specifications
Dell PowerEdge XE9680 at a glance
- Form factor
- 6U rack server, air-cooled
- Processors
- Two 5th Gen Intel Xeon Scalable (up to 64 cores) or 4th Gen Intel Xeon Scalable (up to 56 cores) processors
- Memory
- 32 DDR5 RDIMM slots, up to 4 TB, up to 5600 MT/s (5th Gen) or 4800 MT/s (4th Gen)
- Accelerators
- 8 × NVIDIA HGX H200 141 GB or H100 80 GB SXM5 with NVLink · 8 × AMD Instinct MI300X 192 GB OAM with Infinity Fabric · 8 × Intel Gaudi 3 128 GB OAM with embedded RoCE
- Storage
- Up to 8 × 2.5" NVMe/SAS/SATA SSD (122.88 TB) or up to 16 × E3.S NVMe
- RAID & boot
- PERC H965i (not with Gaudi 3) · BOSS-N1 with 2 × M.2 SSD (RAID 1)
- Expansion
- Up to 10 × PCIe Gen5 x16 slots (8 with Gaudi 3) · 1 × OCP 3.0
- Networking
- 2 × 1 GbE embedded · 6 × 800 Gb OSFP embedded (Gaudi 3 only)
- Power
- 2800 W Titanium hot-swap redundant (200–240 VAC) · 3200 W Titanium (277 VAC, US/Canada) · 3000 W Titanium (Gaudi 3)
- Cooling
- Air cooling with up to 6 mid-tray and up to 10 rear HPR Gold fans (12 with Gaudi 3)
- Management
- iDRAC9, iDRAC Direct, Redfish RESTful API, OpenManage Enterprise, CloudIQ
- Security
- Silicon Root of Trust, signed firmware, Secure Boot, data-at-rest encryption, TPM 2.0 FIPS
- Operating systems
- Ubuntu Server LTS, Red Hat Enterprise Linux, SUSE Linux Enterprise, VMware ESXi
- Dimensions & weight
- H 263.2 mm × W 482 mm × D 1008.77 mm (with bezel) · up to 114.05 kg
Specifications per Dell Technologies. Available accelerators and options vary by region and may be subject to export regulations.
Download Dell spec sheet (PDF)
Why Forte-Tech
Your local partner for Dell infrastructure
We help banks, telecom operators, government entities and enterprises specify, source and deploy Dell PowerEdge servers.
Genuine Dell hardware
Factory-configured PowerEdge servers with Dell manufacturer warranty.
Pre-sales engineering
Engineers who size CPU, memory, storage and power for your actual workload.
Delivery across Egypt & GCC
Logistics to your data center or branch, with optional rack installation.
Tender & PO ready
Formal quotations, compliance documents and invoicing for procurement teams.
How ordering works
Tell us your workloadPick a build or send us your requirements or BOM.
Get a validated quoteOur engineers validate compatibility and send a detailed quotation with pricing and lead time.
Approve & orderPurchase order, bank transfer or tender documentation.
Delivery & deploymentDelivered to your site in Egypt or the GCC, with optional installation.
FAQ
Questions about the XE9680
What is the lead time for the XE9680?
XE9680 systems are built to order, and lead time depends on accelerator availability. We confirm the delivery date with your quote.
Which accelerator should I choose?
Choose the NVIDIA H200 for the widest software support and large-model training, the H100 for proven training and fine-tuning, the AMD MI300X when you need the most GPU memory per server, and Intel Gaudi 3 for Ethernet-based scale-out AI. Tell us your models and number of users and we'll recommend the right platform.
What data-center infrastructure does the XE9680 need?
The XE9680 is an air-cooled 6U server weighing up to 114 kg. It needs a rack and floor rated for its weight, high-power circuits for its 2800 W or 3200 W Titanium power supplies, and strong data-center cooling. Our engineers review your site before delivery.
Is deployment included?
Dell requires ProDeploy Plus or Dell Customer Deployment services for XE-series servers, including GPU subsystem testing. We include the required deployment in your quotation.
What is the difference between the XE9680 and the R770?
The R770 is a 2U server with up to two GPUs, ideal for inference in a standard rack. The XE9680 is a 6U, 8-GPU platform for model training and large-scale inference.
Are there export restrictions on these GPUs?
Some high-end accelerators are subject to U.S. export regulations. We confirm availability and any licensing requirements for your country with your quotation.
What warranty is included?
Every server is supplied with Dell's manufacturer warranty. Extended Dell ProSupport or ProSupport Plus coverage can be added, and the exact warranty terms are stated on your quotation.
Do you deliver outside Egypt?
Yes. We supply customers across Egypt and the GCC, including Saudi Arabia, the UAE, Qatar, Kuwait, Oman and Bahrain.
Planning an AI cluster?
Tell us your models, users and timeline. Our engineers will size the XE9680 system, networking and data-center requirements.